Correlational and thermodynamic properties of finite-temperature electron liquids in

the hypernetted-chain approximation

Shigenori Tanaka

Citation: The Journal of Chemical Physics 145, 214104 (2016); doi: 10.1063/1.4969071
View online: http://dx.doi.org/10.1063/1.4969071
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/145/21?ver=pdfcov
Published by the AIP Publishing

Articles you may be interested in

Probing the structural and dynamical properties of liquid water with models including non-local electron
correlation
J. Chem. Phys. 143, 054506 (2015); 10.1063/1.4927325

Impacts of enhanced electronic correlation in anion p-orbitals on electronic structure and magnetic properties
of nitrogen or carbon doped zinc oxide
J. Appl. Phys. 111, 07E313 (2012); 10.1063/1.3672090

Order-disorder phase transition in two-dimensional Ising model with exchange and dipole interactions
J. Appl. Phys. 99, 08F708 (2006); 10.1063/1.2173209

Jacob’s ladder of density functional approximations for the exchange-correlation energy

AIP Conf. Proc. 577, 1 (2001); 10.1063/1.1390175

Finite-temperature full random-phase approximation model of band gap narrowing for silicon device
simulation
J. Appl. Phys. 84, 3684 (1998); 10.1063/1.368545
 THE JOURNAL OF CHEMICAL PHYSICS 145, 214104 (2016)

Correlational and thermodynamic properties of finite-temperature

electron liquids in the hypernetted-chain approximation
Shigenori Tanakaa)
Graduate School of System Informatics, Kobe University, 1-1 Rokkodai, Nada, Kobe 657-8501, Japan
(Received 28 September 2016; accepted 15 November 2016; published online 6 December 2016)

Correlational and thermodynamic properties of homogeneous electron liquids at finite temperatures

are theoretically analyzed in terms of dielectric response formalism with the hypernetted-chain (HNC)
approximation and its modified version. The static structure factor and the local-field correction
to describe the strong Coulomb-coupling effects beyond the random-phase approximation are self-
consistently calculated through solution to integral equations in the paramagnetic (spin unpolarized)
and ferromagnetic (spin polarized) states. In the ground state with the normalized temperature θ = 0,
the present HNC scheme well reproduces the exchange-correlation energies obtained by quantum
Monte Carlo (QMC) simulations over the whole fluid phase (the coupling constant rs ≤ 100), i.e.,
within 1% and 2% deviations from putative best QMC values in the paramagnetic and ferromag-
netic states, respectively. As compared with earlier studies based on the Singwi-Tosi-Land-Sjölander
and modified convolution approximations, some improvements on the correlation energies and the
correlation functions including the compressibility sum rule are found in the intermediate to strong
coupling regimes. When applied to the electron fluids at intermediate Fermi degeneracies (θ ≈ 1), the
static structure factors calculated in the HNC scheme show good agreements with the results obtained
by the path integral Monte Carlo (PIMC) simulation, while a small negative region in the radial distri-
bution function is observed near the origin, which may be associated with a slight overestimation for
the exchange-correlation hole in the HNC approximation. The interaction energies are calculated for
various combinations of density and temperature parameters ranging from strong to weak degeneracy
and from weak to strong coupling, and the HNC values are then parametrized as functions of r s and
θ. The HNC exchange-correlation free energies obtained through the coupling-constant integration
show reasonable agreements with earlier results including the PIMC-based fitting over the whole fluid
region at finite degeneracies in the paramagnetic state. In contrast, a systematic difference between
the HNC and PIMC results is observed in the ferromagnetic state, which suggests a necessity of
further studies on the exchange-correlation free energies from both aspects of analytical theory and
simulation. Published by AIP Publishing. [http://dx.doi.org/10.1063/1.4969071]

I. INTRODUCTION is essential for constructing the input exchange-correlation

Recently, there has been an increasing interest in the ther-
modynamic properties of uniform electron gas or liquid sys-
tems at finite Fermi degeneracies, which play a pivotal role in
warm dense matter (WDM)1–4 characterized by elevated tem-
peratures and wide compression ranges. A lot of astrophysical
objects and materials under extreme experimental or environ-
mental conditions are associated with the WDM, such as those
highly compressed states observed in compact stars, planet
cores, inertial confinement fusion, and laser ablation.5 Theo-
retical descriptions by ab initio computer simulations for these
materials are important both for interpreting experimental
results and for obtaining insights into those parameter regions
difficult to access experimentally, which could be performed
in the framework of finite-temperature density functional the-
ory (DFT).3,4 In this formulation of finite-temperature DFT
calculations, the accurate information on thermodynamics of
electron fluid over a wide range of density and temperature
a) Electronic address: tanaka2@kobe-u.ac.jp
potentials.
Substantial progress has been achieved since the 1980’s in
the theoretical study of the correlational and thermodynamic
properties of uniform electron gas or liquid at finite tempera-
tures.5 The dielectric and thermodynamic functions have been
calculated first in the random-phase approximation (RPA)6–9
and next in the approaches involving the local-field correc-
tion (LFC)5,10 which describes the strong-coupling effects
beyond the RPA. The latter includes those self-consistent inte-
gral equation approaches based on the parametrized LFC,11 the
Singwi-Tosi-Land-Sjölander (STLS) approximation,12–14 the
dynamical STLS approximation,15 the Vashishta-Singwi (VS)
approximation,16,17 and the modified convolution approxima-
tion (MCA).10,18–20 Through comparison with the existing
Monte Carlo simulation data in the classical and ground-
state limits, these analytic approaches have been expected
to provide fairly accurate predictions for the thermodynamic
functions at any Fermi degeneracy of electrons, e.g., within
several % deviations in the exchange-correlation free ener-
gies from the “true” values in the cases of STLS and MCA
approximations.13,14,20

0021-9606/2016/145(21)/214104/11/$30.00 145, 214104-1 Published by AIP Publishing.

214104-2 Shigenori Tanaka
On the other hand, quantum Monte Carlo (QMC) simula-
tions are in principle expected to provide the exact solution
to the correlational and thermodynamic quantities of elec-
tron gas.21 As for the finite-temperature uniform electron gas,
Brown et al.22 first applied the restricted path integral Monte
Carlo (RPIMC) method in coordinate space to the evalua-
tion of correlational and thermodynamic functions, in which
the fermion nodes of density matrix were fixed at those of
ideal Fermi gas to avoid the sign problem.23 Brown et al.24
also derived an analytic parametrization for the exchange-
correlation free energy as a function of density and temperature
of electron gas on the basis of their RPIMC data. Unfortu-
nately, their simulation results have been found to contain some
systematic errors especially in the low-temperature and high-
density regimes, and the parametrized expression for ther-
modynamic functions shows unphysical behaviors due to its
irrelevant functional form. Karasiev et al. (KSDT)25 then pro-
posed an improved fitting formula for the exchange-correlation
free energy by performing appropriate interpolation with cor-
rect asymptotic limits. In addition, other path integral Monte
Carlo (PIMC) calculations, the permutation blocking path inte-
gral Monte Carlo (PB-PIMC) approach with a higher-order
factorization of density matrix,26,27 and the configuration path
integral Monte Carlo (CPIMC) approach formulated in the
Fock space of Slater determinants27,28 were also carried out
to extend the validity of QMC calculations to wider parame-
ter regions. However, the difficulties associated with the finite
system size (typically 33 particles for spin-polarized case), the
fermion sign problem, and the finite time slice still remain in
spite of continuing efforts to overcome or correct them. (As
for the latest advance in the finite-size correction, see a recent
work by Dornheim et al.29 )
Considering the current situation that a completely
dependable formula for the exchange-correlation free energy
of an electron fluid over any combination of number density n,
temperature T, and spin polarization ζ has not been obtained,
the present study performs novel integral-equation-based cal-
culations for the correlational and thermodynamic proper-
ties of finite-temperature electron liquids on the basis of the
hypernetted-chain (HNC) approximation, which is expected
to give better correlation and thermodynamic functions than
the earlier approaches13,14,20 with comparable accuracy to
the simulation results. The HNC approximation is known to
provide very accurate descriptions for the correlational and
thermodynamic properties of classical one-component plasma
(OCP) over the whole fluid region.5,10,30–32 We here take into
account the HNC approximation in the framework of the
dielectric, density-response formalism at finite temperatures5
which would allow for accurate descriptions both in the weak-
coupling (Hartree-Fock and RPA) and strong-coupling (nearly
crystalline) regimes. We thus calculate the normalized inter-
action energies per electron in units of the Coulomb energy
(e2 /a; see below) over the wide range of density and tem-
perature in the fluid-phase, spin unpolarized (paramagnetic)
and polarized (ferromagnetic) states. In addition, parametrized
expressions for the normalized interaction energies and the
exchange-correlation free energies are derived, which would
be accurate over the whole fluid region of finite-temperature
(including the ground-state and classical limits) electron gas
J. Chem. Phys. 145, 214104 (2016)
or liquid, supposing the smooth interpolation for the normal-
ized interaction energies. Besides, it is noted that the present
HNC scheme on the basis of the dielectric formulation with the
LFC5,10,33 is different from the classical-map HNC (CHNC)
approach34–37 in which the correspondence to the classical
system is utilized and also different from the Fermi-HNC
approach38,39 in which the variational principle with respect
to the ground-state energy is employed.
In the following, the theoretical framework of this work is
first illustrated in Sec. II. Calculational results are then shown
in Sec. III with associated discussions. Conclusions and some
additional remarks are given in Sec. IV.
II. THEORY
A. Physical parameters
Let us consider a homogeneous electron fluid with the
average number density n = N/V and the absolute temperature
T with N and V being the particle number and the volume. The
Coulomb coupling constant for the degenerate electron gas or
liquid is given by5,10
rs = a/aB , (1)
where a = (3/4πn) 1/3
is the Wigner-Seitz radius and
aB = ~2 /me2 is the Bohr radius with ~, m, and −e being the
Planck constant, the mass, and electric charge of an elec-
tron, respectively. In the high-temperature, classical limit, the
strength of Coulomb coupling can instead be measured by5,10
e2
Γ= , (2)
akB T
where k B refers to the Boltzmann constant. The degree of
Fermi degeneracy is then measured by5,10
θ = kB T /EF , (3)
1/3
where EF = ~2 kF2 /2m is the Fermi energy with kF = (3π 2 n)
being the Fermi wavenumber. It is noted that the common E F
1/3
with k F (not with kF0 = (6π 2 n) ) is employed both for the
paramagnetic and ferromagnetic states in the present study,
thus θ is determined only in terms of T and n. There is a useful
relation among the three parameters, r s , Γ, and θ as
Γθ = 2λ2 rs (4)
with λ = (4/9π) . Moreover, the degree of spin polarization
1/3
in the electron fluid is measured by20
ζ = (n1 − n2 )/n, (5)
where n1 and n2 represent the number densities of spin-up and
spin-down electrons, respectively, hence ζ = 0 for the param-
agnetic (spin unpolarized) state and ζ = 1 for the ferromagnetic
(spin polarized) state.
B. Density response formalism
We start with the fluctuation-dissipation relation between
the wavenumber (k)- and frequency (ω)-dependent, retarded
density response function χ+ (k, ω) and the dynamic structure
factor S(k, ω) as5,13
!
~ ~ω
S(k, ω) = − coth Im χ+ (k, ω). (6)
2π 2kB T
214104-3 Shigenori Tanaka J. Chem. Phys. 145, 214104 (2016)

The static structure factor is then obtained through the principal with
n
value (℘) integral over the frequency of S(k, ω) as χ0 (k, 0) = − . (15)
∞ ! kB T
~ ~ω The Ornstein-Zernike relation between the pair correla-
S(k) = − ℘ dω coth Im χ+ (k, ω)
2πn −∞ 2kB T tion function h(r) and the direct correlation function c(r) in
kB T X
∞ classical simple liquids,30
=− χ(k, zl ), (7) 
n l=−∞
h(r) = c(r) + n dr0c(r − r0)h(r0), (16)
where the contributions from the discrete frequencies
zl = 2πilkB T/~ (l = 0, ±1, ±2, . . .) on the imaginary axis have is expressed as
been summed after the deformation of the integral contour.13 ĥ(k) = ĉ(k) + nĉ(k)ĥ(k) (17)
The complex-frequency-dependent, causal density response
function is expressed in the density response formalism as5 in the k-space, where we employ the Fourier transformation
χ0 (k, z) as 
χ(k, z) = , (8) ĥ(k) = dreik·r h(r) (18)
1 − 3(k) [1 − G(k, z)] χ0 (k, z)
where 3(k) = 4πe2 /k 2 and the free-particle polarizability of for the pair correlation function and the analogous expression
electrons, for the direct correlation function. Since the pair correlation
XX fσ (q) − fσ (q + k) function is related to the static structure factor as
χ0 (k, ω) = , (9)
σ q
~ω + (q) − (q + k) + iδ S(k) = 1 + nĥ(k), (19)
have been introduced with (q) = ~2 q2 /2m and δ → +0; the we find a relation,
Fermi distribution function for each spin species σ (up and 3(k) [1 − G(k)] = −kB T ĉ(k), (20)
down)
) −1 through comparison between Eqs. (13) and (17).
(q)
( " #
fσ (q) = exp − ασ + 1 (10) The HNC equation for classical simple liquids reads30
kB T
with the dimensionless chemical potential α σ satisfies the
" #
V (r)
h(r) + 1 = exp − + h(r) − c(r) (21)
normalization condition kB T
X
nσ = fσ (k) (11) with V (r) = e2 /r in the case of electron liquid. This equation
k is rewritten as
with n =
P
nσ , where we use a conventional notation, "  #
σ V (r)
X  dk c(r) = exp − + n dr c(r − r )h(r )
0 0 0
kB T
= . (12) 
(2π)3
k − n dr0c(r − r0)h(r0) − 1 (22)
If we set G(k,z) = 0 in Eq. (8), we recover the random-
phase approximation (RPA) expression for the density with the use of the Ornstein-Zernike relation, Eq. (16). We
response function. The local-field correction (LFC) G(k,z) here take the gradient of r over both the sides of this equation,
thus accounts for the strong-coupling effects beyond the thus finding
RPA5,10 and plays an essential role to accurately describe 
∇V (r)
the correlational and thermodynamic properties of electron ∇c(r) = − [h(r) + 1] + nh(r) dr0∇c(r − r0)h(r0). (23)
kB T
liquids.
Fourier transformation of this equation and the operation of ik
C. Hypernetted-chain (HNC) approximation then lead to33
" #
In this subsection we derive a relation between the LFC 3(k) 1 X
k ĉ(k) +
2
=− k · qĥ(k − q)3(q)
and the static structure factor in the HNC approximation, kB T kB T q
thus providing a closure equation to determine them self- X
consistently in the wavenumber (k) space. Hereafter, we rely +n k · qĥ(k − q)ĥ(q)ĉ(q). (24)
q
on the static approximation to the LFC, i.e., G(k, z) ' G(k),
and the static G(k) is related to S(k) on the basis of classical With the aid of Eqs. (19) and (20), we thus obtain
liquid theory. 1Xk·q 
G(k) = −

Let us note a relation between the static structure factor S(k − q) − 1
n q q2
and the static density response function as10,30
   
kB T × 1 − G(q) − 1 S(q) − 1 . (25)
S(k) = − χ(k, 0) (13)
n This LFC form in the HNC approximation is reduced
in the classical limit (θ → ∞). The static density response to two well-known expressions, the STLS and convolution
function is then written as approximation (CA) types,10 as limiting cases. If we set G(q)
χ0 (k, 0) = 1 on the right-hand side of Eq. (25), we recover the STLS
χ(k, 0) = (14) expression of LFC. On the other hand, if we set G(q) = 0, we
1 − 3(k) [1 − G(k)] χ0 (k, 0)
214104-4 Shigenori Tanaka J. Chem. Phys. 145, 214104 (2016)

recover the CA expression. Considering this property, we here is fully quantum-mechanical up to the approximation level
divide the LFC into two parts as of RPA. As for the description of the strong-coupling effect
beyond the RPA, we rely on the classical HNC approximation
G(k) = G1 (k) + G2 (k) (26)
for the static LFC G(k).
with
1Xk·q  A. Classical limit
G1 (k) = −

2
S(k − q) − 1 (27)
n q q
First, we consider the classical limit, in which Eqs. (13)–
and (15) are used for the calculation of S(k). It is known that
1Xk·q  the HNC approximation accurately describes the correlational
G2 (k) =
  
S(k − q) − 1 G(q) − 1 S(q) − 1 . and thermodynamic properties of the classical electron gas or
n q q2
the one-component plasma (OCP) in the whole fluid regime
(28)
(Γ . 200).5,10 Figure 1 illustrates the radial distribution func-
G1 (k) represents the STLS part of LFC, and as seen in
tion (RDF) at Γ = 10 calculated by
Sec. III, G2 (k) refers to a small correction to G1 (k) in the
long-wavelength region. 1X
g(r) = 1 + h(r) = 1 + [S(k) − 1] e−ik·r (33)
In addition to the HNC approximation above, another n
k
approximation called the modified hypernetted-chain (MHNC)
approximation is examined in the present study. On the basis from the static structure factor S(k) obtained in the HNC
of an analogous idea to the MCA18–20 in which some improve- approximation mentioned above. This solution is identical to
ments on the correlational and thermodynamic properties over that obtained in terms of the usual way of solving the classical
the CA have been observed, S(k − q) in Eq. (28) is replaced HNC integral equation set,31,32 and shows a better agreement
with a screening function defined by with the Monte Carlo (MC) simulation results40 than the corre-
k2 sponding STLS and MHNC results shown in the same figure.
S(k) = , (29) The interaction energy per particle is then calculated as
k 2 + ks2 
Eint n  1X
so that the second term of LFC in the MHNC approximation = drV (r) g(r) − 1 =

3(k) [S(k) − 1] ; (34)
is given by N 2 2
k

1Xk·q f g
G2 (k) = the HNC approximation is known31,32 to reproduce the MC
 
2
S(k − q) − 1 G(q) − 1 S(q) − 1 .
n q q values40,41 within the errors of 1% in the whole fluid region in
(30) contrast to the STLS approximation which shows 5.0% under-
Performing the angular integration in Eq. (30), we obtain the estimation of the magnitude of E int at Γ =10 as compared to
expression for G2 (k) over a one-dimensional integral as the MC value.19,42,43
 ∞  The LFCs calculated at Γ = 10 are compared in Fig. 2
ks2 q2 + k 2 + ks2  (q − k)2 + ks2 
 
G2 (k) = +

dq 1 ln among the HNC, MHNC, and STLS results. In the limit of
(2π)2 n 0  (q + k) + ks 
 4kq  2 2 
 k → 0, G(k) is proportional to k 2 and the proportionality coef-
ficient should be consistent with that evaluated from the ther-
  
× G(q) − 1 S(q) − 1 . (31)
modynamic function (compressibility sum rule, see below).
The screening wavenumber k s is then determined self-
The thermodymanic42,43 asymptotic behavior of G(k) is also
consistently via a constraint as
illustrated in Fig. 2, showing the improvement of agreement
1X f g 1X
3(k) S(k) − 1 = 3(k) [S(k) − 1] , (32)
2 2
k k

which implies that the interaction energy calculated from S(k)

is identical to that from S(k). Since G1 (k) can also be calculated
in terms of a one-dimensional integral, the total LFC in the
MHNC approximation is expressed over the one-dimensional
integration in the wavenumber space, which markedly reduces
the computational cost and well retains the numerical stability
concerning the convergence of solution to integral equations
for G(k) and S(k). The MHNC solution to G(k) can also be used
as the initial input for solving the HNC integral equations.

III. RESULTS AND DISCUSSION

On the theoretical basis explained in Sec. II, we can obtain
G(k) and S(k) through a self-consistent solution to the integral
equations, Eqs. (7), (8), and (25), in the HNC approximation.
In the case of MHNC approximation, Eq. (30) is employed
for the G2 (k) part instead of Eq. (28). The present approach
FIG. 1. Radial distribution function (RDF) at Γ = 10 in the classical limit.
Blue solid line, green dotted line, and red dashed line represent the HNC,
MHNC, and STLS results, respectively. Black filled circles refer to the Monte
Carlo (MC) data.40
214104-5 Shigenori Tanaka J. Chem. Phys. 145, 214104 (2016)

system in the ground state (T = 0). In this case the summation

over the discrete integers l in Eq. (7) is converted into the inte-
gration over the continuous variable z. Figures 3(a) and 3(b)
illustrate the radial distribution functions (RDFs) calculated
in the HNC and MHNC approximations for the paramagnetic
state (ζ = 0) at r s = 1 and 10, respectively. Compared with
g(r) obtained through the diffusion Monte Carlo (DMC) sim-
ulation,44 both the HNC and MHNC results show a fairly good
overall agreement, but somewhat lower values and small nega-
tive region (at r s = 10) near the origin (r = 0) since the positivity
of g(r) in the present HNC scheme is not guaranteed except for
the classical limit. It is also noted that the extent of the nega-
tive region of g(r) is reduced in comparison to that in the MCA
scheme.20
The compressibility sum rule for the LFC is expressed
by
FIG. 2. Local field correction (LFC) at Γ = 10 in the classical limit. Blue solid
line, green dotted line, and red dashed line represent the HNC, MHNC, and
∂ 2 Fxc
!
STLS results, respectively. Purple dotted line refers to the asymptotic curve
obtained from the MC equation of state.42,43 lim [−3(k)G(k)] = 2 , (35)
k→0 ∂n V
in the cases of HNC and MHNC over the STLS result (22.2, 
where Fxc V refers to the exchange-correlation free energy per
10.6, and 39.3 overestimations compared to the MC thermo-
volume. Figure 4 compares the compressibility coefficient for
dynamic value42,43 at Γ = 10 in the cases of HNC, MHNC,
the LFC,
and STLS approximations, respectively). This improvement
on the compressibility sum rule is due to the inclusion of the !2
kF
 
correction term G2 (k) in the LFC (see also Fig. 5 below). γLFC = lim  G(k) , (36)
k→0  k 
B. Ground state (T = 0) 

The present HNC scheme has also been applied to the spin obtained in the HNC, MHNC, and STLS calculations with that
unpolarized (ζ = 0) and fully polarized (ζ = 1) electron gas calculated from the equation of state,

FIG. 3. RDF in the ground state (T = 0). Blue solid line and green dotted line represent the HNC and MHNC results, respectively. Black filled circles refer to
1/3
the result obtained by the diffusion Monte Carlo (DMC) simulation.44 (a) Calculated results at r s = 1 for the paramagnetic state (ζ = 0), where kF = (3π 2 n) .
1/3
(b) Calculated results at r s = 10 for the paramagnetic state (ζ = 0). (c) Calculated results at r s = 10 for the ferromagnetic state (ζ = 1), where kF = (6π n) .
0 2
214104-6 Shigenori Tanaka J. Chem. Phys. 145, 214104 (2016)

Through numerical integrations of ε int (rs , θ) calculated

FIG. 4. Compressibility coefficient γ for 0.1 ≤ rs ≤ 100 at T = 0 and ζ = 0.
Blue solid line, green dotted line, and red dashed line represent the HNC,
MHNC, and STLS results for G(k), respectively. Purple dotted line refers to
the values calculated from the exchange-correlation energies45 fitted to the
Green’s function Monte Carlo (GFMC) data.21
kF2∂ 2 Fxc
!
γEOS = − , (37)
4πe2 ∂n2 V
where the exchange-correlation energies for the ground-state
paramagnetic electron gas21 parametrized as a function of r s
by Vosko et al.45 are employed. We thus observe that the
self-consistency concerning the compressiblity sum rule is
substantially improved in the HNC and MHNC schemes over
the STLS scheme as in the classical limit. Figure 5 illustrates
the division of the LFC into the G1 (k) and G2 (k) parts as in
Eq. (26) in the HNC scheme at r s = 10 and ζ = 0, showing
that the G2 (k) works as a long-wavelength correction to the
STLS-type LFC.
The interaction energy per particle in units of the Coulomb
energy, ε int (rs , θ) = Eint /N(e2 /a), is calculated from the
correlation function in terms of Eq. (34). The exchange-
correlation free energy per particle in units of the Coulomb
energy, fxc (rs , θ) = Fxc /N(e2 /a), is then evaluated through the
coupling-constant integration of ε int (rs , θ) as5,20
 metallic densities. In the strong-coupling regime (rs & 10)
1 rs
fxc (rs , θ) = dx ε int (x, θ). (38)
rs 0
in the HNC and MHNC schemes for rs ≤ 100 (data on
30 points shown in the supplementary material), we have
obtained the values of the exchange-correlation energy
ε xc (that is, fxc at T = 0) in the paramagnetic and ferromag-
netic states, which are listed in Tables I and II, respectively. In
these tables the ε xc values obtained by various quantum Monte
Carlo (QMC) calculations21,44,46 are also listed in addition to
the HNC, MHNC, MCA,20 and STLS13,14 values. As seen
in Table I, the HNC values of ε xc in the paramagnetic state
somewhat underestimate (overestimate in the magnitude) the
QMC values, but the deviations from the best (i.e., lowest) val-
ues of GFMC or DMC are less than 1% over the whole fluid
region (rs . 100). Considering that the VMC and fixed-node
DMC calculations give upper bounds to the exact ground-state
energy, this feature in the HNC approximation seems fairly
well. The STLS (and MHNC) calculations show very good
agreement with the QMC results at metallic densities of elec-
trons (2 . rs . 6), which is due to a fortunate cancellation
of errors (see below). In the case of the ferromagnetic state,
on the other hand, the deviations of the HNC values of ε xc
from the QMC values amount to about 2% for 4 . rs . 8.
This underestimation of ε xc in the HNC approximation can
be attributed to the overestimation of the exchange-correlation
hole in the short-range regime when two electrons with parallel
spins approach each other. This defect in the HNC approx-
imation may be observed in Fig. 3(c), where the RDF at
r s = 10 in the ferromagnetic state shows a deeper negative
region around the origin than in the paramagnetic state. It
is supposed that the HNC scheme describes the correlation
hole due to the Coulomb repulsion very accurately, as seen
in the classical limit (Fig. 1), but when the exchange (Pauli)
hole predominates, its description for the total exchange-
correlation hole would lead to some overcounting. This would
also explain the reason why the STLS approximation, which
describes the correlation hole with reduced degree, gives an
accurate evaluation for the exchange-correlation energies at
near the Wigner crystallization, in contrast, the correlation

TABLE I. Negative of exchange-correlation energy per particle in units of

e2 /a, −εxc , for various values of r s in the paramagnetic state (ζ = 0) at
T = 0. The HNC, MHNC, MCA,20 and STLS13,14 results are compared with
the variational Monte Carlo (VMC),46 the diffusion Monte Carlo (DMC),44,46
and the Green’s function Monte Carlo (GFMC)21 results. The symbol “· · · ”
means that the corresponding values are not available.

rs VMC46 DMC46 DMC44 GFMC21 HNC MHNC MCA STLS

0.5 ··· ··· 0.495 ··· 0.499 0.498 0.498 0.497

1.0 0.511 0.514 0.517 0.518 0.522 0.521 0.521 0.519
2.0 0.542 0.546 0.548 0.548 0.553 0.552 0.552 0.548
3.0 0.567 0.570 0.570 ··· 0.575 0.573 0.574 0.569
4.0 0.583 0.589 ··· ··· 0.592 0.589 0.592 0.585
5.0 0.596 0.599 0.600 0.599 0.606 0.603 0.606 0.598
8.0 0.626 0.630 ··· ··· 0.636 0.631 0.637 0.626
10.0 0.641 0.646 0.646 0.644 0.650 0.645 0.652 0.639
20.0 ··· ··· 0.691 0.688 0.693 0.685 0.697 0.678
FIG. 5. Division of the LFC G(k) into the G1 (k) and G2 (k) parts calculated in 50.0 ··· ··· ··· 0.743 0.743 0.730 0.748 0.721
the HNC approximation at T = 0, r s = 10, and ζ = 0. Red dashed line, green 100.0 ··· ··· ··· 0.777 0.774 0.756 0.779 0.746
dotted line, and blue solid line refer to G1 (k), G2 (k), and G(k), respectively.
214104-7 Shigenori Tanaka J. Chem. Phys. 145, 214104 (2016)

TABLE II. Negative of exchange-correlation energy per particle in units of

e2 /a, −εxc , for various values of r s in the ferromagnetic state (ζ = 1) at
T = 0. The HNC, MHNC, MCA,20 and STLS13,14 results are compared with
the VMC,46 DMC,44,46 and GFMC21 results. The symbol “· · · ” means that
the corresponding values are not available.

rs VMC46 DMC46 DMC44 GFMC21 HNC MHNC MCA STLS

0.5 ··· ··· 0.596 ··· 0.600 0.600 0.600 0.598

1.0 0.604 0.606 0.608 ··· 0.613 0.613 0.613 0.611
2.0 0.619 0.622 0.624 0.625 0.632 0.631 0.632 0.628
3.0 0.632 0.633 0.637 ··· 0.646 0.644 0.646 0.641
4.0 0.640 0.643 ··· ··· 0.656 0.654 0.658 0.651
5.0 0.650 0.651 0.654 0.654 0.665 0.663 0.667 0.659
8.0 0.669 0.671 ··· ··· 0.685 0.681 0.688 0.676
10.0 0.679 0.681 0.683 0.682 0.695 0.690 0.698 0.685
20.0 ··· ··· 0.714 0.713 0.724 0.717 0.730 0.711
50.0 ··· ··· ··· 0.755 0.761 0.749 0.768 0.740 FIG. 6. Static structure factors calculated in the HNC approximation at θ = 1
100.0 ··· ··· ··· 0.785 0.785 0.768 0.791 0.758 and ζ = 0. Blue, purple, green, red, and orange curves refer to the results for
S(k) at r s = 0.1, 1, 10, 40, and 100, respectively.

effect becomes dominant and the HNC description becomes
superior to the STLS one. It is also noted that the effect by
the negative region of g(r) around r ∼ 0 on the interaction
energy E int is much smaller than the effect due to the inaccu-
racy of S(k) in the long-wavelength (k → 0) region because of
the contrast between V (r) ∝ 1/r and v(k) ∝ 1/k 2 in Eq. (34)
for the Coulombic system. In order to resolve this problem
about the underestimation of g(r) in the short-range region,
we would need to incorporate the frequency-dependent LFC in
the fully quantum-mechanical framework of strong-coupling
theory.
with γEOS (Eq. (37); see also Eq. (41) below) as well as in the
ground-state case (Fig. 4) in the light of relatively small impact
on G(k) (see Fig. 2).
D. Parametrized equation of state
Using 107 HNC values of ε int calculated for rs ≤ 100
at θ = 0, 0.2, 1, and 5 each in the paramagnetic (ζ = 0) and
ferromagnetic (ζ = 1) states (see the supplementary material),
fitting expressions,

C. Finite temperatures ai (θ) + bi (θ)rs1/2 + ci (θ)rs

We have performed the MHNC and HNC calculations for
the spin unpolarized (ζ = 0) and fully polarized (ζ = 1) elec-
tron liquids at θ = 0.2, 1, and 5 to obtain S(k) and G(k) in the
density range of rs ≤ 100 (totally 77 points at each ζ). The
numerical procedure to obtain the self-consistent solution is
similar to that in the STLS case13,14,47 in which the l summa-
tion in Eq. (7) was retained up to |l| = 1000. We first carried
out the MHNC calculations starting with a small value of r s
at each θ and using the convergent G(k) as the initial value
for the calculation at larger r s . The HNC calculations were
next performed using the MHNC solution to G(k) as the initial
guess at the same values of θ and r s . The RDF and the inter-
action energy were then calculated according to Eqs. (33) and
(34). The HNC and MHNC data for ε int are compiled in the
supplementary material.
Figure 6 illustrates the HNC results of the static structure
factor S(k) for various values of r s at θ = 1 in the paramagnetic
state (ζ = 0). These results are compared well with the PIMC
results22 in Fig. 7, indicating a good accuracy of the HNC
approximation in the intermediate to long wavelength regions.
The HNC and MHNC results for RDF g(r) at θ = 1, r s = 10, and
ζ = 0 are depicted and compared to the PIMC result in Fig. 8,
which show a small negative region near the origin analogous
to that in the ground state (see Fig. 3). The estimated values
of compressibility coefficient γLFC , Eq. (36), in the HNC and
MHNC approximations are depicted in Fig. 9 as functions of
r s at θ = 1 and ζ = 0, showing a reasonable comparison
ε int (rs , θ, i) = − , (39)
1 + di (θ)rs1/2 + ei (θ)rs
have been constructed as functions of r s and θ, where
i = 0 and 1 refer to ζ = 0 and 1, respectively. The coefficients
ai − ei in Eq. (39) are parametrized as functions of θ in a
universal form as
1 + x2 θ 2 + x3 θ 3 + x4 θ 4
f (θ) = F(θ) , (40)
1 + y2 θ 2 + y3 θ 3 + y4 θ 4
where the values of x j and yj ( j = 2,3,4) and the functional
forms of F(θ) are tabulated in Table III, and the form of the
exchange (Hartree-Fock) contribution ai (θ) is employed from
the expression proposed by Perrot and Dharma-wardana.8 This
fitting formula, which has been constructed with the Monte
Carlo optimization, reproduces the 107 HNC values of ε int in
the paramagnetic state with the mean absolute relative errors
of 0.18% and the maximum relative errors of 1.50%; in the
case of the ferromagnetic state, these errors are 0.15% and
0.94%, respectively. In the classical limit (θ → ∞), the expres-
sions Eqs. (39) and (40) reproduce the HNC values31,32 for
Γ ≤ 200 with digressions of less than 0.74%, and are reduced
to the Debye-Hückel form5,10 in the weak-coupling limit
(Γ → 0).
The expression for the exchange-correlation free energy
fxc is then obtained by performing the coupling-constant
integration as prescribed in Eq. (38). We thus find
214104-8 Shigenori Tanaka J. Chem. Phys. 145, 214104 (2016)

FIG. 7. Comparison of the static structure factors calculated in the HNC approximation and by the path integral Monte Carlo (PIMC) simulation22 for various
values of r s at θ = 1 and ζ = 0. (a) Blue solid curve and green dotted curve refer to the HNC results at r s = 1 and 10, respectively; red and orange filled circles
refer to the PIMC results at r s = 1 and 10, respectively. (b) Results at r s = 40, where the blue solid curve, green dotted curve, and red filled circles refer to the
HNC, MHNC, and PIMC results, respectively.

! " ! !#
ci 2 ci di −1/2 1 ci di ci di
fxc (rs , θ, i) = − − bi − rs − ai − − bi − ln |ei rs + di rs1/2 + 1|
ei ei ei ei rs ei ei ei
di2 1/2
+ di 
! !      
2 ci ci di   −1  2ei rs di

+  
di ai − e + 2 − e  − tan−1
 
1/2
 * + bi −  tan   1/2 

1/2  
 (41)
2
ei (4ei − di ) rs  i i - e i 
 
 2
 (4ei − di )  
(4e i − d 2 ) 
i
,   


for i = 0 (paramagnetic state) and 1 (ferromagnetic state). The
normalized exchange-correlation free energies fxc in the HNC
approximation are illustrated in Fig. 10 as functions of r s .
In Fig. 10(a) for the paramagnetic state at θ = 1, the HNC
values of fxc for 0.01 ≤ rs ≤ 100 are compared with the
values obtained by other methods such as the MCA,20 the
STLS approximation,13,14 the improved parametrization based
on the STLS values (iSTLS),48 the Vashishta-Singwi (VS)
approximation,17 the classical-map HNC (CHNC) approx-
imation,34,35 and the PIMC22,24 based parametrization by
Karasiev et al. (KSDT).25 It is observed in Fig. 10(a) that the
overall agreements among all the fitting formulas are fairly
good, except for the digressions by the VS approximation and
the CHNC approximation for 1 . rs . 10. It is noted here
that the iSTLS parametrization48 was performed to improve
the defects in the original STLS-based parametrization13,14 in
the strongly degenerate (θ  1) regime and in the strong-
coupling (rs , Γ  1) regime with the aid of known asymptotic
behaviors.5 The CHNC results may contain some inaccuracies
for S(k) in the intermediate to long wavelength regions.
Very recently, after the present work was complete,
an improved result for fxc on the basis of finite-size cor-
rected (FSC) QMC (PB-PIMC and CPIMC) simulations has
been reported.29 Figure 10(b) compares the results of fxc by
the HNC and KSDT parametrizations with the FSC-QMC
based fitting in the paramagnetic state as the functions of

FIG. 9. Compressibility coefficient γ for 0.01 ≤ rs ≤ 100 at θ = 1 and

ζ = 0. Blue solid line and green dotted line represent the HNC and MHNC
FIG. 8. RDF at θ = 1 and r s = 10 in the paramagnetic state (ζ = 0). Blue solid results obtained via Eq. (36) from G(k), respectively. Purple dotted line refers
line and green dotted line represent the HNC and MHNC results, respectively. to the values calculated via Eq. (37) from the fitting formula for the exchange-
Black filled circles refer to the result obtained by the PIMC simulation.22 correlation free energies (Eq. (41)).
214104-9 Shigenori Tanaka J. Chem. Phys. 145, 214104 (2016)

TABLE III. Values of x j and yj ( j = 2,3,4) and functional forms of F(θ) in Eq. (40) for the coefficients ai − ei (i = 1, 2) appearing in Eq. (39). The symbol
“· · · ” means that the corresponding terms have not been adopted in the approximations.

f (θ) F(θ) x2 x3 x4 y2 y3 y4

a0 (θ) 0.458 165 tanh(1/θ) 4.058 17 0.123 027 2.271 33 8.310 51 ··· 5.1105
b0 (θ) 0.529 835d0 (θ) 13.397 7 2.742 82 ··· 8.256 68 2.234 50 ···
c0 (θ) 0.860 254e0 (θ) 0.987 103 0.369 317 ··· 1.048 62 0.353 894 ···
d0 (θ) 0.547 950 tanh(1/θ 1/2 ) 6.564 92 126.406 1.790 76 1.0 75.698 2 1.0
e0 (θ) 0.399 854 tanh(1/θ) 1.302 60 0.495 918 1.964 24 1.0 1.105 77 1.0
a1 (θ) 1.259 92a0 (0.629 961θ) ··· ··· ··· ··· ··· ···
b1 (θ) 0.607 303d1 (θ) 5.020 70 1.759 51 ··· 3.387 78 1.644 93 ···
c1 (θ) 0.855 862e1 (θ) 0.757 902 0.357 122 ··· 0.802 781 0.340 461 ···
d1 (θ) 0.547 458 tanh(1/θ 1/2 ) 2.683 86 134.934 1.794 46 1.0 75.914 4 1.0
e1 (θ) 0.338 480 tanh(1/θ) 1.871 25 1.905 19 2.282 95 1.0 0.157 029 1.0

r s at θ = 1, 4, and 8, where the latter QMC simulations results. In contrast to the paramagnetic case at θ = 1, the
were carried out for θ ≥ 0.5 and 0.1 ≤ rs ≤ 10. It is HNC and MCA results deviate from the KSDT results more
observed in this figure that the HNC results overestimate the substantially. In the weak-coupling (i.e., high-density) regime
magnitude of fxc  by about 2% compared to the FSC-QMC (rs . 1), there remains a slight deviation up to r s = 0.01. In
results for 1 . rs . 10 at θ = 1. On the other hand, the strong-coupling regime (1 . rs . 100), the difference
the KSDT parametrization overestimates the fxc  values of
FSC-QMC by 7%–10% for 0.1 . rs . 1 at θ = 4 and 8.
As for the ferromagnetic state, the HNC results for fxc
at θ = 1 are compared in Fig. 10(c) with the MCA,20 the
CHNC,34,35 and the PIMC-based parametrization (KSDT)25
(e.g., 5.5% at r s = 10 between the HNC and KSDT results)
is larger than expected from the comparisons between the
HNC and simulation results in the classical and degenerate
limits. Even considering the fact that the HNC approximation
may underestimate the QMC values of fxc by about 2% in the

FIG. 10. Exchange-correlation free energies per particle in units of e2 /a, fxc , as functions of r s . (a) Paramagnetic state (ζ = 0) at θ = 1; the HNC, MCA,20
STLS,13,14 iSTLS,48 Vashishta-Singwi (VS),17 classical-map HNC (CHNC),34,35 and PIMC-based fitting (KSDT)25 results are represented by the blue solid
line, pink dashed line, red dashed line, green dotted line, yellow-green solid line, orange solid line, and purple dotted line, respectively. (b) Comparison of the
HNC and KSDT results with the finite-size corrected (FSC) QMC results29 for the paramagnetic state at θ = 1, 4, and 8; blue, green, and red curves refer to the
HNC, KSDT, and FSC-QMC results, respectively; solid, dashed, and dotted lines represent the results for θ = 1, 4, and 8, respectively. (c) Ferromagnetic state
(ζ = 1) at θ = 1, where it is noted that “θ = 1” in the present study corresponds to t = 2−2/3 = 0.629 961 in the KSDT fitting25 for the ferromagnetic state; the
HNC, MCA, CHNC, and KSDT results are represented by blue solid line, pink dashed line, orange filled circles, and purple dotted line, respectively.
214104-10 Shigenori Tanaka J. Chem. Phys. 145, 214104 (2016)

ferromagnetic state for θ → 0 and also some errors in the fitting a number of proposed expressions for fxc thus appear to con-
procedures, we may thus propose that the re-examination on verge well to a common (probably correct) evaluation in the
the exchange-correlation free energies for the spin-polarized paramagnetic state. On the other hand, even considering the
electron gas system at finite temperatures should be per- expected 2% – 3% underestimation for fxc (overestimation for
formed from both sides of QMC simulations and analytic
approaches.
We have thus derived the analytic expressions for ε int
and fxc as the functions of θ and r s both in the param-
agnetic (ζ = 0) and ferromagnetic (ζ = 1) states. In order
to obtain the expressions available at any spin polarization
(0 < ζ < 1), we can use the interpolation scheme developed
in the MCA study.20 It is expected that the MCA interpola-
tion function which depends on ζ, r s , and θ would work well
also in the HNC case, because the differences in ε int and fxc
between the HNC and MCA approximations are very minor
over the whole rs − θ regions and the functional dependence
on the degree of spin polarization is expected to be smooth and
simple.
IV. CONCLUDING REMARKS
The present work has performed the calculations for
the correlational and thermodynamic functions of finite-
temperature electron liquids in the paramagnetic and ferro-
magnetic states over a wide range of density and temperature
parameters on the basis of the HNC and MHNC approxima-
tions. Primary attention has been focused upon the construc-
tion of accurate equations of state for homogeneous electron
fluid which can be used as the input for finite-temperature
DFT calculations with good accuracy comparable to those
obtained through QMC simulations. The accuracies for the
exchange-correlation free energies achieved in the present
HNC study are superior to those in the CHNC34,35 and VS17
schemes, as seen in Fig. 10(a). As compared to the earlier
STLS results,13,14 the improvements of the consistency in
the compressibility sum rule and of the correlation energy in
the strong-coupling regime have been observed. Further, the
improvements of the RDF in the short-range region and of the
correlation energy in the intermediate-coupling regime have
been found over the MCA results.20 The accurate knowledge
of G(k) is essential for constructing the dielectric screening
function of electrons (k, ω)5,10 and the associated static struc-
ture factor S(k); the information about S(k) has recently been
utilized to perform the finite-size correction for QMC ther-
modynamic data.29 The information on g(r) or S(k) of the
uniform electron gas can also provide the key input to the
finite-temperature DFT calculations for inhomogeneous sys-
tems in which the density-gradient corrections may be taken
into account.49
The parametrized expression for the exchange-correlation
free energy obtained in the present HNC study is expected to
reproduce the exact values of fxc with the relative errors of
less than a few % over the whole n − T region of the fluid
phase in the paramagnetic and ferromagnetic states. The prob-
able digressions in the intermediate-coupling regime would be
ascribed to the small negative region of g(r) near the origin,
but the influence of this error upon thermodynamic functions
is minor in comparison to that associated with S(k) in the long-
wavelength region. Through comparison with earlier studies,
fxc ) by the present HNC scheme, the differences between the
HNC and PIMC results in the intermediate-coupling (1 . rs
. 10) and weak-coupling (rs . 1) regimes observed at θ = 1
in the ferromagnetic state (Fig. 10(c)) seem to be too large, thus
suggesting the necessity of further QMC studies and associated
re-examination on the equation of state.
SUPPLEMENTARY MATERIAL
See supplementary material for the interaction energies
of electron liquids at θ = 0, 0.2, 1, and 5, ζ = 0 and 1, and
rs ≤ 100 calculated in the HNC and MHNC approximations.
ACKNOWLEDGMENTS
The author would like to acknowledge the Grants-in-
Aid for Scientific Research (Grant No. 26460035) from the
Ministry of Education, Culture, Sports, Science and Tech-
nology (MEXT). The numerical computations in the present
work were carried out by the IBM eServer p7 model 755
at the Information Science and Technology Center of Kobe
University.
1 S. P. Regan, K. Falk, G. Gregori, P. B. Radha, S. X. Hu, T. R. Boehly,
B. J. B. Crowley, S. H. Glenzer, O. L. Landen, D. O. Gericke, T. Döppner,
D. D. Meyerhofer, C. D. Murphy, T. C. Sangster, and J. Vorberger, Phys.
Rev. Lett. 109, 265003 (2012).
2 L. B. Fletcher, A. L. Kritcher, A. Pak, T. Ma, T. Döppner, C. Fortmann,
L. Divol, O. S. Jones, O. L. Landen, H. A. Scott, J. Vorberger, D. A. Chap-
man, D. O. Gericke, B. A. Mattern, G. T. Seidler, G. Gregori, R. W. Falcone,
and S. H. Glenzer, Phys. Rev. Lett. 112, 145004 (2014).
3 M. W. C. Dharma-wardana, Computation 4, 16 (2016).
4 V. V. Karasiev, L. Calderin, and S. B. Trickey, Phys. Rev. B 93, 063207
(2016).
5 S. Ichimaru, H. Iyetomi, and S. Tanaka, Phys. Rep. 149, 91 (1987).
6 U. Gupta and A. K. Rajagopal, Phys. Rev. A 22, 2792 (1980).
7 U. Gupta and A. K. Rajagopal, Phys. Rep. 87, 259 (1982).
8 F. Perrot and M. W. C. Dharma-wardana, Phys. Rev. A 30, 2619 (1984).
9 D. G. Kanhere, P. V. Panat, A. K. Rajagopal, and J. Callaway, Phys. Rev. A
33, 490 (1986).
10 S. Ichimaru, Rev. Mod. Phys. 54, 1017 (1982).
11 R. G. Dandrea, N. W. Ashcroft, and A. E. Carlsson, Phys. Rev. B 34, 2097
(1986).
12 K. S. Singwi, M. P. Tosi, R. H. Land, and A. Sjölander, Phys. Rev. 176, 589
(1968).
13 S. Tanaka and S. Ichimaru, J. Phys. Soc. Jpn. 55, 2278 (1986).
14 S. Tanaka, S. Mitake, and S. Ichimaru, Phys. Rev. B 32, 1896 (1985).
15 H. K. Schweng and H. M. Böhm, Phys. Rev. B 48, 2037 (1993).
16 P. Vashishta and K. S. Singwi, Phys. Rev. B 6, 875 (1972).
17 T. Sjostrom and J. Dufty, Phys. Rev. B 88, 115123 (2013).
18 K. Tago, K. Utsumi, and S. Ichimaru, Prog. Theor. Phys. 65, 54 (1981).
19 X.-Z. Yan and S. Ichimaru, J. Phys. Soc. Jpn. 56, 3853 (1987).
20 S. Tanaka and S. Ichimaru, Phys. Rev. B 39, 1036 (1989).
21 D. M. Ceperley and B. J. Alder, Phys. Rev. Lett. 45, 566 (1980).
22 E. W. Brown, B. K. Clark, J. L. DuBois, and D. M. Ceperley, Phys. Rev.
Lett. 110, 146405 (2013).
23 D. M. Ceperley, J. Stat. Phys. 63, 1237 (1991).
24 E. W. Brown, J. L. DuBois, M. Holzmann, and D. M. Ceperley, Phys. Rev.
B 88, 081102(R) (2013).
25 V. V. Karasiev, T. Sjostrom, J. Dufty, and S. B. Trickey, Phys. Rev. Lett.
112, 076403 (2014).
26 T. Dornheim, T. Schoof, S. Groth, A. Filinov, and M. Bonitz, J. Chem. Phys.
143, 204101 (2015).
214104-11 Shigenori Tanaka
27 S. Groth, T. Schoof, T. Dornheim, and M. Bonitz, Phys. Rev. B 93, 085102
(2016).
28 T. Schoof, S. Groth, J. Vorberger, and M. Bonitz, Phys. Rev. Lett. 115,
130402 (2015).
29 T. Dornheim, S. Groth, T. Sjostrom, F. D. Malone, W. M. C. Foulkes, and
M. Bonitz, Phys. Rev. Lett. 117, 156403 (2016).
30 J.-P. Hansen and I. R. McDonald, Theory of Simple Liquids, 3rd ed.
(Academic Press, London, 2006).
31 J. F. Springer, M. A. Pokrant, and F. A. Stevens, Jr., J. Chem. Phys. 58, 4863
(1973).
32 K.-C. Ng, J. Chem. Phys. 61, 2680 (1974).
33 Ph. Choquard, in Strongly Coupled Plasmas, edited by G. Kalman (Plenum,
New York, 1978), p. 347.
34 M. W. C. Dharma-wardana and F. Perrot, Phys. Rev. Lett. 84, 959
(2000).
35 F. Perrot and M. W. C. Dharma-wardana, Phys. Rev. B 62, 16536
(2000).
36 S. Dutta and J. Dufty, Europhys. Lett. 102, 67005 (2013).
37 Y. Liu and J. Wu, J. Chem. Phys. 141, 064115 (2014).
38 J. L. Lantto, Phys. Rev. B 22, 1380 (1980).
J. Chem. Phys. 145, 214104 (2016)
39 J. G. Zabolitzky, Phys. Rev. B 22, 2353 (1980).
40 S. G. Brush, H. L. Sahlin, and E. Teller, J. Chem. Phys. 45, 2102
(1966).
41 J.-P. Hansen, Phys. Rev. A 8, 3096 (1973).
42 W. L. Slattery, G. D. Doolen, and H. E. DeWitt, Phys. Rev. A 21, 2087
(1980).
43 W. L. Slattery, G. D. Doolen, and H. E. DeWitt, Phys. Rev. A 26, 2255
(1982).
44 G. G. Spink, R. J. Needs, and N. D. Drummond, Phys. Rev. B 88, 085121
(2013).
45 S. H. Vosko, L. Wilk, and M. Nusair, Can. J. Phys. 58, 1200 (1980).
46 G. Ortiz and P. Ballone, Phys. Rev. B 50, 1391 (1994).
47 The convergence criterion adopted in the self-consistent integral equations
for S(k) and G(k) at θ = 0, 0.2, 1, and 5 is ∫ 0xmax dx[Gout (x) − Gin (x)]2 < δ
for the input and output LFC values in the iteration process, where x = k/k F
(or k/kF0 ), x max = 1000.0, δ = 10−8 for rs < 1.0, and δ = 10−8 rs for rs ≥ 1.0.
48 S. Ichimaru, Rev. Mod. Phys. 65, 255 (1993); S. Tanaka, “Improved equation
of state for finite-temperature spin-polarized electron liquids on the basis of
Singwi-Tosi-Land-Sjolander approximation” (unpublished).
49 J. P. Perdew and Y. Wang, Phys. Rev. B 46, 12947 (1992).